I'm not sure why civitai is not detecting my lora for remixing, add the lora manually if you want to remix.

V3 -> V4: Increased Dim/Alpha to 32, significantly improved tagging, "on stomach" works more consistently now and will generate massive bellies, if you find the sizes too big, lower LoRA strength. Recommended samplers: DPM++ 2M Karras 40 steps, Euler a 30 steps. Usage of adetailer with inpainting denoise strength of 0.6 is highly recommended.

V2 -> V3: Cleaned up the dataset and re-tagged all the training files. Girls are generally bigger, and "on stomach" work almost all the time. I am providing both network dim/alpha =16 and 8 since 16 has some over fitting issues. Try both and pick your favorite. Again, use DPM++ 3M SDE Karras or Euler A, DPM++ 2M Karras performs pretty bad on this lora.

V1 -> V2: Doubled number of training steps, bellies are generally bigger, use "on stomach" to let a girl lie on her belly. Reduce LORA strength if the bellies are too big or the images are overbaked.

Make a girl's belly big, no questions asked. Belly size is a bit uncontrollable atm, but you will always get bigger than natural belly sizes. Can increase breasts sizes as well.

Use DPM++ 3M SDE Karras with around 40 steps, always use adetailer.

If DPM++ 3M SDE Karras is not available, USE Euler A not DPM++ 2M Karras!

No longer applicable for V4.